How we got there: let him walk

We asked: when a team signs another team's free agent at big money, how often does the production survive? The public contract_history view carries every contract since 2005 with sequence-derived flags: contract_no, and the rung (early-extension, at-expiry-resign, external).

The result (scripted; construction in the article backtest block): retention = per-game production two seasons after vs two before; bust = keeping less than half. Incumbent deals bust at 6 to 7 percent regardless of rung; external deals at 13.0 percent (gap 6.5 points, 90 percent CI 4.4 to 8.7). The stack: a career-year spike the incumbent re-signed busts 9.4 percent; the same spike signed externally busts 33.3 percent.

Reproduce the sample: filter contract_history to contract_no >= 2, year_signed 2013 to 2023, and top-quartile apy_cap_pct within a position. The retention join runs over player_season and defense_player_season by player_id and signing-year windows; a coverage-quality outcome for corners is the known gap (tackle-based outcomes cannot see a corner's collapse). Fork the chart above to explore the contract market itself.